A multi-verse optimizer-based CNN-BiLSTM pixel-level detection model for peanut aflatoxins.

Area of Science:

• Agricultural Science
• Food Science
• Computational Science

Background:

• Peanuts are prone to aflatoxin contamination, a serious health risk.
• Accurate, real-time detection of aflatoxins is crucial for food safety.

Purpose of the Study:

• To improve pixel-level aflatoxin detection accuracy in hyperspectral images.
• To develop an optimized deep learning model for precise aflatoxin identification.

Main Methods:

• A Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) fusion model was developed.
• The model was optimized using the Multi-Verse Optimizer (MVO) algorithm.
• Fine-tuning was performed using hyperspectral data of aflatoxins at various concentrations.

Main Results:

• The MVO-optimized CNN-BiLSTM model achieved 94.92% validation accuracy and 95.59% recall.
• This model outperformed traditional machine learning (SVM, AdaBoost) and other deep learning methods (CNN, CNN-LSTM).
• Accuracy improvements ranged from 3.08% to 6.93% compared to existing methods.

Conclusions:

• The MVO-CNN-BiLSTM model significantly enhances pixel-level aflatoxin detection accuracy.
• This advancement supports the development of effective online monitoring systems for aflatoxins.
• The findings contribute to improved food safety and public health by enabling better detection of peanut contamination.